6  Metrics Overview

We explore how WiFi gives us a snapshot of urban life. It’s about identifying where people are (Location), counting them (Count), tracking their movements (Track), recognizing repeat visits (Revisits), and understanding their actions (Activities). This framework extends the human-sensing taxonomy that Teixeira et al. (2010) propose in their Figure 2 (presence, count, location, track, and identity). Presence is subsumed by Count, since any device counted is a device present. We rename identity to Revisits, reflecting a privacy-preserving scope focused on device-level visit patterns rather than personal identification, and add an Activities metric that captures stationary behavior. The graphic below illustrates these concepts step by step:

Overview of WiFi sensing metrics: from pinpointing locations to understanding urban activities. Adapted and extended from Figure 2 of Teixeira et al. (2010).

Location: Where Are People at a Specific Time?

‘Location’ is determined by analyzing the characteristics of WiFi signals at a given time, such as signal frequency and strength. Depending on the situation, we might assign locations based on sensor positions or estimate them in areas between sensors.

Before counting, we must pinpoint location. Without knowing where devices are, we just have a total count.

Imagine a room with three people, just like in the diagram above:

  • Without location data, counting detected device MAC addresses only tells you there are three devices total.
  • With even rough location data, you can determine that two devices are on the left and one on the right, providing much richer information.

Count: How Many Over Time?

‘Count’ focuses on tallying unique MAC addresses to estimate how many people are in an area over different timescales. This method allows for long-term monitoring of device presence, offering insights into occupancy trends over extended periods.

Track: Where and When Did They Move?

‘Track’ involves recording when and where phones are detected. This data lets us follow people’s movements, highlighting the dynamic patterns of city life and offering insights into how different areas are used throughout the day.

Revisits: Who Comes Back?

With ‘Revisits’, we analyze data over time to recognize returning visitors and understand their movement habits. This metric provides insights into recurring traffic patterns and the behaviors of different user groups within the city.

Activities: What Are They Doing?

‘Activities’ aims to understand actions by analyzing timing and movement data. We look at how people use different city areas (whether they stay or pass through), observing changes in activity patterns at different times and days.

WarningA Note on MAC Randomization

Modern mobile devices (iOS since 2014, Android since 2015) randomize their MAC addresses for privacy protection. This affects all metrics to varying degrees:

  • Count shifts from overestimation (one device broadcasting many addresses) to a lower bound once randomized addresses are removed in preprocessing (Chapter 5)
  • Track sequences may break when a device changes its MAC mid-journey
  • Revisits classification over extended periods becomes unreliable

These limitations are most significant in uncontrolled public spaces. In controlled environments (campuses with authenticated WiFi, events with registered devices), the impact is reduced. Mitigation strategies are discussed in the Count chapter and in Appendix B.

To assess accuracy, we compare WiFi-derived metrics against ground truth data such as manual counts or GPS logs. The Location chapter validates position estimates against GPS ground truth; the other metrics build on those localized detections.